Abstract:We propose novel methods for Conditional Value-at-Risk (CVaR) estimation for nonlinear systems under high-dimensional dependent random inputs. We develop a novel DD-GPCE-Kriging surrogate that merges dimensionally decomposed generalized polynomial chaos expansion and Kriging to accurately approximate nonlinear and nonsmooth random outputs. We use DD-GPCE-Kriging (1) for Monte Carlo simulation (MCS) and (2) within multifidelity importance sampling (MFIS). The MCS-based method samples from DD-GPCE-Kriging, which is efficient and accurate for high-dimensional dependent random inputs, yet introduces bias. Thus, we propose an MFIS-based method where DD-GPCE-Kriging determines the biasing density, from which we draw a few high-fidelity samples to provide an unbiased CVaR estimate. To accelerate the biasing density construction, we compute DD-GPCE-Kriging using a cheap-to-evaluate low-fidelity model. Numerical results for mathematical functions show that the MFIS-based method is more accurate than the MCS-based method when the output is nonsmooth. The scalability of the proposed methods and their applicability to complex engineering problems are demonstrated on a two-dimensional composite laminate with 28 (partly dependent) random inputs and a three-dimensional composite T-joint with 20 (partly dependent) random inputs. In the former, the proposed MFIS-based method achieves 104x speedup compared to standard MCS using the high-fidelity model, while accurately estimating CVaR with 1.15% error.
What problem does this paper attempt to address?
This paper attempts to solve the problem of estimating the Conditional Value - at - Risk (CVaR) of nonlinear systems under high - dimensional correlated random input conditions. Specifically, the paper focuses on how to calculate CVaR efficiently and accurately, especially when facing high - dimensional and dependent random inputs.
### Problem Background
In many engineering designs, especially in the presence of uncertainties, such as uncertainties in manufacturing processes, material properties or operating environments, risk measures are important tools for ensuring system safety and reliability. As a risk - measure tool, CVaR can reflect the average loss value beyond a certain risk level, so it has been widely used in the financial and engineering fields. However, traditional CVaR estimation methods (such as Monte Carlo simulation) are inefficient when dealing with high - dimensional correlated random inputs, especially when each sample requires expensive high - fidelity model evaluations.
### Main Contributions of the Paper
To solve the above problems, the paper proposes the following innovative methods:
1. **New DD - GPCE - Kriging surrogate model**:
- Combines the Dimensionally Decomposed Generalized Polynomial Chaos Expansion (DD - GPCE) and the Kriging method to approximate highly nonlinear and non - smooth random outputs more accurately.
- By fusing these two techniques, DD - GPCE - Kriging can exhibit better performance when dealing with high - dimensional correlated random inputs.
2. **CVaR estimation methods based on Monte Carlo simulation (MCS) and Multi - Fidelity Importance Sampling (MFIS)**:
- Proposes two CVaR estimation methods based on DD - GPCE - Kriging: one is to directly use DD - GPCE - Kriging for Monte Carlo simulation; the other is to combine multi - fidelity importance sampling.
- For the MCS method, although a large number of samples can be generated efficiently, bias may be introduced due to the dependence on the approximation quality of DD - GPCE - Kriging.
- Therefore, a method based on MFIS is further proposed, which uses DD - GPCE - Kriging to determine the bias density and draws a small number of samples from the high - fidelity model to provide an unbiased CVaR estimate.
3. **Accelerate bias density construction**:
- In order to accelerate the construction of the bias density, the low - fidelity model with lower computational cost is used to calculate DD - GPCE - Kriging.
### Experimental Verification
The paper verifies the effectiveness of the proposed methods through numerical experiments. Specifically:
- Tests are carried out on two - dimensional composite laminates (with 28 partially correlated random inputs) and three - dimensional composite T - joints (with 20 partially correlated random inputs).
- The experimental results show that the MFIS - based method is more accurate than the MCS - based method in the case of non - smooth outputs.
- In the example of two - dimensional composite laminates, the proposed MFIS method achieves a 104 - fold speedup compared to the standard MCS method, while the CVaR estimation error is only 1.15%.
### Summary
The paper proposes a novel computational framework for efficiently and accurately estimating CVaR under high - dimensional correlated random input conditions. By combining DD - GPCE and Kriging and using multi - fidelity importance sampling, this method not only improves computational efficiency but also ensures the accuracy of the estimation results. This provides a powerful tool for risk assessment in complex engineering problems.